[PyTorch] ReflectionPad2d illegal memory access

[ssnl]

Python 3.6.4 |Anaconda, Inc.| (default, Jan 16 2018, 18:10:19)
[GCC 7.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> m = torch.nn.ReflectionPad2d(2)
>>> x = torch.randn(1,1,2,2)
>>> m(x)
=================================================================
==6656==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x6020000056e8 at pc 0x7f8316f2e3b8 bp 0x7ffe6c372690 sp 0x7ffe6c372688
READ of size 4 at 0x6020000056e8 thread T0
    #0 0x7f8316f2e3b7 in .omp_outlined..521 (/opt/conda/lib/python3.6/site-packages/torch/lib/libATen.so+0x16c43b7)
    #1 0x7f832258dac2 in __kmp_invoke_microtask (/opt/conda/bin/../lib/libiomp5.so+0xb5ac2)
    #2 0x7f832255c6ec in __kmp_fork_call (/opt/conda/bin/../lib/libiomp5.so+0x846ec)
    #3 0x7f83225347dd in __kmpc_fork_call (/opt/conda/bin/../lib/libiomp5.so+0x5c7dd)
    #4 0x7f8316edc880 in .omp_outlined..365 (/opt/conda/lib/python3.6/site-packages/torch/lib/libATen.so+0x1672880)
    #5 0x7f832258dac2 in __kmp_invoke_microtask (/opt/conda/bin/../lib/libiomp5.so+0xb5ac2)
    #6 0x7f832255c256 in __kmp_invoke_task_func (/opt/conda/bin/../lib/libiomp5.so+0x84256)
    #7 0x7f832255d497 in __kmp_fork_call (/opt/conda/bin/../lib/libiomp5.so+0x85497)
    #8 0x7f83225347dd in __kmpc_fork_call (/opt/conda/bin/../lib/libiomp5.so+0x5c7dd)
    #9 0x7f8316edbccd in THNN_FloatSpatialReflectionPadding_updateOutput (/opt/conda/lib/python3.6/site-packages/torch/lib/libATen.so+0x167
1ccd)
    #10 0x7f8315e0146e in at::CPUFloatType::reflection_pad2d_forward(at::Tensor const&, at::ArrayRef<long>) const (/opt/conda/lib/python3.6
/site-packages/torch/lib/libATen.so+0x59746e)
    #11 0x7f8319077d95 in torch::autograd::VariableType::reflection_pad2d_forward(at::Tensor const&, at::ArrayRef<long>) const /var/lib/jen
kins/workspace/torch/csrc/autograd/generated/VariableType.cpp:14878:39
    #12 0x7f8315ff8140 in at::Type::reflection_pad2d(at::Tensor const&, at::ArrayRef<long>) const (/opt/conda/lib/python3.6/site-packages/t
orch/lib/libATen.so+0x78e140)
    #13 0x7f8319073fc1 in torch::autograd::VariableType::reflection_pad2d(at::Tensor const&, at::ArrayRef<long>) const /var/lib/jenkins/wor
kspace/torch/csrc/autograd/generated/VariableType.cpp:14833:23
    #14 0x7f8319d548e3 in at::reflection_pad2d(at::Tensor const&, at::ArrayRef<long>) /var/lib/jenkins/workspace/torch/lib/tmp_install/incl
ude/ATen/Functions.h:2499:29
    #15 0x7f8319d548e3 in torch::autograd::dispatch_reflection_pad2d(at::Tensor const&, at::ArrayRef<long>) /var/lib/jenkins/workspace/torc
h/csrc/autograd/generated/python_nn_functions_dispatch.h:423
    #16 0x7f8319cd88ec in torch::autograd::THPVariable_reflection_pad2d(_object*, _object*, _object*) /var/lib/jenkins/workspace/torch/csrc
/autograd/generated/python_nn_functions.cpp:694:19
    #17 0x563e63b1efd3 in _PyCFunction_FastCallDict (/opt/conda/bin/python3.6+0x10efd3)
    #18 0x563e63bacbeb in call_function (/opt/conda/bin/python3.6+0x19cbeb)
    #19 0x563e63bd1199 in _PyEval_EvalFrameDefault (/opt/conda/bin/python3.6+0x1c1199)
    #20 0x563e63ba59a5 in _PyEval_EvalCodeWithName (/opt/conda/bin/python3.6+0x1959a5)
    #21 0x563e63ba6a10 in fast_function (/opt/conda/bin/python3.6+0x196a10)
    #22 0x563e63baccc4 in call_function (/opt/conda/bin/python3.6+0x19ccc4)
    #23 0x563e63bd1199 in _PyEval_EvalFrameDefault (/opt/conda/bin/python3.6+0x1c1199)
    #24 0x563e63ba6e4a in _PyFunction_FastCallDict (/opt/conda/bin/python3.6+0x196e4a)

Fix incoming.